Online Monitoring

The aim of this project is to investigate fluid-bed processes both experimentally and theoretically. The experimental part of the project deals with the development of an online monitoring system for pharmaceutical manufacturing processes based on Near-Infrared (NIR) spectroscopy. Once proven under laboratory conditions, the monitoring system will be transferred to the production facility of an RCPE industrial partner to demonstrate the capabilities of the system in selected industrial applications. Parallel to developing the monitoring system, a coupled discrete element modeling / computational fluid dynamics (DEM/CFD) simulation tool for fluid-bed processes is going to be developed. By combining particle and fluid dynamics simulations, fluid-bed drying systems can be investigated in level of detail not previously possible with CFD simulations alone. The coupled models will be used to investigate relevant process design issues such as optimal drying and gas-feed strategies, scale-up and reactor design. In addition, the models will also be applied in developing and analyzing concepts for continuous drying systems.

An assessment of NIR systems for appropriate application of process analytical technology (PAT) to monitor granular manufacturing processes was carried out by conducting both, NIR equipment performance (i.e., spectral resolution, sensitivity, measurement time, detector stability, etc.) and measurement probe design (i.e., directly attached spectrometer, single fiber probe, multiple fiber probe systems, etc.). Here, two different systems, an FT-NIR spectrometer and an industrial NIR imaging system, were selected to meet a broad range of experimental demands (e.g., monitoring of very fast processes, monitoring of batch homogeneity with multiple probe systems, etc.). For the spectral data evaluation, multivariate data analysis (MVDA) software packages were introduced for powerful data preprocessing and statistical model building. Thereby a focus was put on dynamic calibration to develop robust qualitative and quantitative models for real-time process monitoring based on experimental design, as well as on batch statistical process control (BSPC) for predefined process trajectories. Currently, these different monitoring principles are experimentally implemented and tested on a fluidized bed process at the Rutgers University to monitor powder particle size and moisture content.

The DEM code, which runs on a single graphics processing unit (GPU), is now able to incorporate two million particles per GigaByte of graphics memory in single-phase operation.

Currently, we achieved a coupled two-phase flow simulation of one million particles in an air stream. The improvement of our developed code will focus on boundary condition handling in the fluid phase as well as on CAD data import. In tandem with the experimental research, the realistic simulation and the understanding of processes in a fluid bed dryer is the aim of the future work in this project.

The application of PAT in pharmaceutical manufacturing processes yields a mechanistic process understanding and enables not only feedback but also feed-forward control of a process to meet product quality for direct release. However, knowledge based selection of critical-to-quality parameters, appropriate equipment for in-line measurements and powerful data management tools have to be developed and implemented for a robust and valid monitoring. For a process monitoring based on spectral data acquisition, like NIR spectroscopy, external influences (e.g., temperature, humidity, etc.) show a strong impact on the monitoring performance, which have to be incorporated according to the product design space. Beside these experimentally challenging issues, data logging for licensed processes and model handling on different levels with univariate (e.g., temperature, pressure, etc.) and multivariate (e.g., spectra, images, etc.) data, have to be correlated and organized through smart meta-programs.

In order to get a better understanding of granular flows numerical simulations are applied. We are using the Discrete Element Method (DEM). There particles are (still) modeled as spheres and their translational and rotational motion are computed due to exerting forces on inter-particle contacts. In the DEM algorithm all particles scan for their neighbors independently. This allows for parallelization. By using cutting edge Nvidia/CUDA technology our unique DEM code can include up to 8 million particles in a simulation.

Suitable extensions for spraying and wetting of powders are developed. This allows studying the influence of liquid-bridges between particles on homogeneity and mixedness in blending process simulations. Here, the calibration of simulation parameters is of great importance. These adjustments require the PAT and NIR data from the experimental project branch, in order to predict process simulation results using DEM.

Numerical simulations of drying in fluidized beds require the coupling of computational fluid dynamics (CFD) and DEM. Our development strategy is to employ the lattice-Boltzmann method (LBM) as CFD solver, because of the numerical robustness and the excellent parallelization features of this technique. A CUDA implementation of an LBM solver is under development.